Setup & Integration

Q: Can dv-search integrate with an existing search solution?

Integration can be as lightweight as required: from browser integration with typeahead, to query expansion and relaxation that can feed an existing search solution, to a full blown search solution with self learning capabilities. The choice depends on your existing needs and can gradually evolve with your business needs and technical capabilities.

Q: What are the pre-requisites?

Data exchanges with your product catalog data is the minimum integration point. Additionally, integration with a product inventory solution can improve the user experience, by dropping out-of-stock products from search results or suggestions. Lastly, integration with our search performance tracking API helps improve product rankings, and self learning.

Q: How long does it typically take to implement dv-search?

From a few days to a few weeks, depending on the use case decided upon.


Q: Can I manually tune rankings?

We do not recommend force ranking products, as it is a highly manual operation that typically interferes with the self learning solution that converges to the best ranking. The display of product or category recommendations alongside search results might be a viable alternative, that combines the benefit of allowing a business to promote certain products based on their needs, while not interfering with the self learning solution that ranks products according to many signals.

Q: How do you deal with product going out of stock?

An integration with an inventory API is typical and allows to drop out-of-stock products from search results or typeahead suggestions.

Q: Do you have upper limits for scaling?

Our solution is fully stateless and can scale horizontally to handle very high volumes. Such capabilities can be demonstrated through synthetic loads in advance of serving real traffic.

Q: What are your SLAs?

Most software responsible for personalization suffer from an over-reliance on historical data and customer profiles instead of real-time intent clues. Our solutions bridge this gap.


Q: How do you test your data updates?

Our data sets are immutable and are versioned. We test rankings continuously using a "gold standard" and can detect changes in search performance rapidly. We use statistical models that significantly limit the risk of introducing negative biases to search performance.

Q: Are data updates optional?

You may opt to freeze the version of data at a point in time when you deem search performance to be satisfactory. However you will be missing out on new updates that are continuously added to our knowledge graph.


Q: Do you have a roadmap?

We continuously innovate and stay on top of all new developments pertaining to search, information retrieval and NLS from both academia and commercial solutions. We do not maintain a public roadmap.

Q: How about features like search-as-you-type?

Search-as-you-type could be implemented today with our existing solution. Talk to us about enabling this use case.

Q: Do you offer visual search?

While image recognition has achieved major improvements in the last few years, detecting and classifying the nuances of images remains difficult, particularly for "soft" objects such as apparel. The most advanced solutions currently make use of 3D photography, which is emerging and may become more widespread in the years to come.

Q: How about voice based search?

Speech recognition is now a mature technology that can be associated with dv-search. The user experience on mobile can be particularly effective, especially with the evolution of NLS and chatbots.


Q: Who are you and where are you located?

We are a small group of engineers and scientists with experience in eCommerce. Among other things, we speak (and our two PhDs also teach) NLS, information retrieval and ranking models, as well C++ and other fast and efficient languages. We are based in San Francisco and Europe.